The incorporation of artificial intelligence (AI) tools into clinical practice is a reality in multiple healthcare settings⁽¹⁾. In Pediatric Emergency Services (SUP), proposals are emerging aimed at reducing the bureaucratic load⁽²⁾. Notably, ambient AI systems are capable of integrating “invisibly” into the consultation room and assisting in the generation of clinical documentation based on the interaction between the professional, the patient, and their family.

Ambient AI groups together solutions that, through ambient microphones, capture the clinical conversation passively and non-intrusively. This signal is processed using advanced language models and subsequently offers assistance to the professional in different tasks; one of the most relevant is the preparation of a structured draft of the episode report.

In some cases, these systems incorporate contextual data from the care process (e.g., access to the clinical history, values of medical vital signs, or results of complementary explorations), broadening their utility and, at the same time, demanding reinforced ethical and governance guarantees (privacy, consent, and data management)⁽³⁾.

In the European environment, any realistic implementation must consider three layers: the protection of personal data under the General Data Protection Regulation, the institutional and contractual governance of processing, and the requirements applicable to the software as a medical device and its CE (Conformité Européenne) marking. Furthermore, the European AI regulatory framework (Regulation (EU) 2024/1689, “AI Act”) establishes specific obligations for certain uses of AI considered high-risk, such as health⁽⁴⁾.

The potential of this technology is evident. A significant portion of the emergency physician’s time is devoted to writing reports, with the consequent impact on clinical presence. This documentary load and interaction with electronic systems have been quantified in different healthcare settings^(5,6). Delegating part of the repetitive work to ambient AI could free up time for activities with greater added value: communication with families, decision-making, team supervision, or coordination with other care facilities. From an organizational point of view, the partial automation of documentation can contribute to improving efficiency, standardizing the structure of reports, and reducing the overload associated with record-keeping. However, this potential only translates into real improvement if it is accompanied by an indispensable principle: all generated content must be reviewed, corrected, and always validated by a pediatrician. AI can assist, but it does not replace clinical responsibility or the obligation to ensure the quality, accuracy, and suitability of the report to the episode attended⁽⁷⁾.

Sin embargo, cuando se plantea la implantación de estos sistemas en hospitales docentes, surge de forma inevitable una cuestión que trasciende lo tecnológico: ¿quién debería tener acceso a esta herramienta dentro de la estructura del hospital? La respuesta no es trivial, porque la realidad y las necesidades de un adjunto no son las mismas que las de un residente en formación.

For pediatric emergency physicians, writing the clinical report largely represents an administrative task added to intense care activity. Prioritization, team coordination, and decision-making rely on already consolidated competencies. In this context, ambient AI can be a resource to regain clinical time and reduce bureaucratic load. Its automation can be considered a reasonable advance, provided that the generated content is reviewed and validated, and demanding standards of safety, quality, and traceability are maintained⁽²⁾.

The scenario is different when considering the use of AI by residents, especially in the early years of training. For a first-year resident (R1), writing the clinical report is not just an administrative obligation; it is an essential part of learning. During their training process, the resident acquires the ability to collect data, discriminate relevant information, order symptoms and findings, develop a differential diagnosis, and justify a therapeutic plan. This process is consolidated when they are forced to make their reasoning explicit: ordering their ideas, detecting information gaps, and deciding what to include and how it is presented. And a significant part of that learning occurs when structuring a coherent clinical report⁽⁸⁾.

Drafting the report forces the explicit articulation of reasoning that might otherwise remain implicit or incomplete. Writing allows for confronting decisions, identifying contradictions, and reviewing the coherence of the approach⁽⁹⁾. If, from the beginning of the residency, AI is the one structuring the information, selecting what is relevant, and drafting the final text, there is a risk that the resident will not fully develop these skills. The reports might have an impeccable appearance but fail to reflect the degree of maturity of the clinical thinking of the person signing them. This can result in training deficits that are not detected until the professional must make decisions with greater autonomy in advanced stages of the training program or in environments with less support⁽¹⁰⁾.

Therefore, access to AI in teaching SUPs should be approached from a phased and pedagogically reasoned perspective. A first phase of implementation limited to pediatric emergency physicians seems prudent: they are the ones in the best position to evaluate quality, identify errors, adjust the system, and establish criteria for safety and adequacy, including outcome indicators and performance audit mechanisms.

In the case of residents, it is necessary to propose differentiated usage models. One option would be to restrict or modulate access during the first year so that the R1 personally prepares their reports in most situations. Another alternative is to have specific modes for teaching, where the AI generates a structured outline (main problems, chronology, key data) that the resident must transform into a complete report. It could also be required that, during certain periods, the report generated by the AI be critically reviewed and compared with the resident’s own draft, using the divergences as a training and supervision opportunity⁽¹¹⁾.

These strategies share one premise: ambient AI should not replace learning but rather integrate as a support tool, preserving the need for the resident to think, argue, and write. In initial phases, it is reasonable for reports generated with the help of AI to be identified (internally) to avoid confusion about the actual degree of professional participation and to facilitate improvement and auditing processes.

The reflection on AI in pediatric emergencies involves considering its impact on specialized training, the quality of clinical reasoning, and patient safety. Decisions about who should use it and under what conditions should be adopted with the participation of teaching commissions, management, and specific working groups (such as those for AI and teaching within the framework of the SEUP) with the aim of defining usage frameworks that balance efficiency and the training process.

AI has the potential to become an important ally for SUPs, helping to reduce the bureaucratic load and improve the care experience for patients, families, and professionals. However, its implementation in the teaching environment requires careful planning and an explicit reflection on its effect on the acquisition of competencies. Ultimately, the challenge is to ensure that technology frees up time without depriving those who are learning of an essential tool for consolidating their clinical thinking. If AI saves us from writing, it should not save us from learning to think.

REFERENCES

1. Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44–56. Available from: https://doi.org/10.1038/s41591-018-0300-7.
2. Verghese A, Shah NH, Harrington RA. What this computer needs is a physician: humanism and artificial intelligence. JAMA. 2018;319(1):19–20. Available from: https://doi.org/10.1001/jama.2017.19198.
3. Martinez-Martin N, Luo Z, Kaushal A, et al. Ethical issues in using ambient intelligence in health-care settings. Lancet Digit Health. 2021;3(2):e115–e123. Available from: https://doi.org/10.1016/S2589-7500(20)30275-2.
4. European Parliament and Council of the European Union. Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Off J Eur Union. 2024.
5. Sinsky C, Colligan L, Li L, et al. Allocation of physician time in ambulatory practice: a time and motion study. Ann Intern Med. 2016;165(11):753–760. Available from: https://doi.org/10.7326/M16-0961.
6. Hill RG Jr, Sears LM, Melanson SW. 4000 clicks: a productivity analysis of electronic medical records in a community hospital emergency department. Am J Emerg Med. 2013;31(11):1591–1594. Available from: https://doi.org/10.1016/j.ajem.2013.06.028.
7. Cabitza F, Rasoini R, Gensini GF. Unintended consequences of machine learning in medicine. JAMA. 2017;318(6):517–518. Available from: https://doi.org/10.1001/jama.2017.7797.
8. Eva KW. What every teacher needs to know about clinical reasoning. Med Educ. 2005;39(1):98–106. Available from: https://doi.org/10.1111/j.1365-2929.2004.01972.x.
9. Charlin B, Tardif J, Boshuizen HPA. Scripts and medical diagnostic knowledge: theory and applications for clinical reasoning instruction. Acad Med. 2000;75(2):182–190. Available from: https://doi.org/10.1097/00001888-200002000-00020.
10. Norman G, Monteiro S, Sherbino J. The causes of errors in clinical reasoning: cognitive biases, knowledge deficits, and dual process thinking. Acad Med. 2017;92(1):23–30. Available from: https://doi.org/10.1097/ACM.0000000000001421.
11. Masters K. Artificial intelligence in medical education. Med Teach. 2019;41(9):976–980. Available from: https://doi.org/10.1080/0142159X.2019.1595557.